* possible macvlan packet loss
@ 2023-01-24 15:14 Aleksander Dutkowski
2023-01-24 23:45 ` Vadim Fedorenko
2023-01-25 1:44 ` Jakub Kicinski
0 siblings, 2 replies; 3+ messages in thread
From: Aleksander Dutkowski @ 2023-01-24 15:14 UTC (permalink / raw)
To: netdev
Hello,
we are experiencing something, that seems to be packet loss on macvlan
interface.
Input is ~350Mbps of multicasts, scattered over 120 docker containers,
each having macvlan on top of the same mellanox 10G NIC. Each of
macvlans have bcqueuelen set to INT_MAX.
Streams recorded from phys interface have all packets, whereas streams
recorded from macvlan have drops.
I wonder what is the best way to test and measure the queue fulfillment?
We can do simple printf when the queue is full, but maybe there are
some tools or techniques I'm not aware of that we can use?
Thanks
Alex
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: possible macvlan packet loss
2023-01-24 15:14 possible macvlan packet loss Aleksander Dutkowski
@ 2023-01-24 23:45 ` Vadim Fedorenko
2023-01-25 1:44 ` Jakub Kicinski
1 sibling, 0 replies; 3+ messages in thread
From: Vadim Fedorenko @ 2023-01-24 23:45 UTC (permalink / raw)
To: Aleksander Dutkowski; +Cc: Linux Kernel Network Developers
On 24.01.2023 15:14, Aleksander Dutkowski wrote:
> Hello,
>
> we are experiencing something, that seems to be packet loss on macvlan
> interface.
> Input is ~350Mbps of multicasts, scattered over 120 docker containers,
> each having macvlan on top of the same mellanox 10G NIC. Each of
> macvlans have bcqueuelen set to INT_MAX.
> Streams recorded from phys interface have all packets, whereas streams
> recorded from macvlan have drops.
>
Is it real packet loss or just reordering of the packets that is accounted by
the software as a loss?
> I wonder what is the best way to test and measure the queue fulfillment?
> We can do simple printf when the queue is full, but maybe there are
> some tools or techniques I'm not aware of that we can use?
>
> Thanks
> Alex
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: possible macvlan packet loss
2023-01-24 15:14 possible macvlan packet loss Aleksander Dutkowski
2023-01-24 23:45 ` Vadim Fedorenko
@ 2023-01-25 1:44 ` Jakub Kicinski
1 sibling, 0 replies; 3+ messages in thread
From: Jakub Kicinski @ 2023-01-25 1:44 UTC (permalink / raw)
To: Aleksander Dutkowski; +Cc: netdev
On Tue, 24 Jan 2023 16:14:03 +0100 Aleksander Dutkowski wrote:
> I wonder what is the best way to test and measure the queue fulfillment?
> We can do simple printf when the queue is full, but maybe there are
> some tools or techniques I'm not aware of that we can use?
In terms of debugging tools bpftrace (potentially combined with kprobes)
will change your life.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2023-01-25 1:44 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2023-01-24 15:14 possible macvlan packet loss Aleksander Dutkowski
2023-01-24 23:45 ` Vadim Fedorenko
2023-01-25 1:44 ` Jakub Kicinski
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).